A Deep Learning Model Approach Using Feature Engineering To Predict Melanoma Tumour Size

Melanoma, a lethal ailment, occurs when the melanocytes in our body become cancerous. The fatality rate for early detection of melanoma is relatively low, making early diagnosis critical. While most studies focus on classification techniques to identify the presence of malignant melanoma, this paper suggests a novel deep learning approach to estimate tumour size quantitatively. Initially, the features are pre-processed using a square root transformation function to improve the quality of the dataset, followed by the addition of novel features. These features are fed to an Artificial Neural Network to predict tumour size. This study compares the model performance before and after the addition of handcrafted features for different optimization algorithms. Excellent performance is obtained, with a very low mean square error of 0.0001 and a high coefficient of determination of 0.9976 for an Adam-optimized model using feature construction for the development set of data.